[funkybomber]

Hello everyone!

I just need your advice on troubleshooting my folding rig (BOINC).

My rig:

Motherboard: Gigabyte GA-P55A-UD6
CPU: Core-i5 750
GPU 1: ATI HD5850
GPU 2: AMD HD7870XT
PSU: 500W

And here's the full story...

For a long time I have been folding with CPU only, because many of the GPU projects that are available for my 5850 just didn't interest me
that much (for example Collatz Conjecture).
[email protected] did interest me, but for some reason my gpu would not complete tasks successfuly.

All that changed a couple of weeks ago when I decided to switch to drivers Catalyst v13.1. From that point on my 5850 was working fine.

A couple of days ago I decided to give some extra "umph" to my folding rig, by adding another GPU. The sweet spot GPU seemed to be the AMD 7870XT so I just went and bought it.

I switched the cards and put the 7870XT at PCI-E slot 1 (at the place where the 5850 used to be) and the 5850 at PCI-E slot 2.

I run BOINC after I edited the cc_config.xml to set the "use_all_gpus" value to "1". BOINC did indeed recognise both cards and started crunching. However, after a couple of minutes I started to see artifacts on my desktop, especially during browsing. Tasks seemed to complete successfuly though.

Checking for the latest Catalyst drivers was out of the question since I had read that they ruin support for [email protected]

So I just decided to do the silliest thing. I switched the cards again, content to sacrifice the preformance of the 7870XT as my main GPU just to have it folding quietly on the background, hoping at least that the 5850 would not show these artifacts.

After the switch, things seemed to work as I expected for a good deal of time (4-5 hours). However, something started bothering me. Some tasks started taking WAY too much time to complete. For example, a task that would need 2m20s on my 5850 would complete in 15minutes or so, with the GPU apparently on full load (according to the GPU-Z).

I thought that it might be a couple of bad tasks, but the situation became permanent. The 7870XT crunching efficiently and the 5850 crunching inefficiently.

Last thing I know is that my 7870XT is not crunching at all anymore, and my 5850 continues to crunch inefficiently.

Could it be that the PSU is not pumping enough juice to keep both of them GPUs going? But then again, shouldn't have the whole system just crashed?

As a last resort, I tried to underclock both GPUs' memory clocks in hope of reducing the energy the need (as the memory apparently doesn't contribute anything to Milkyway crunching). Also I disabled the crunching on my CPU in order to reduce power consumption further. None of that seemes to do the trick.

I'm open to suggestions guys, what should I try next?

Thanks in advance.

 

[freelian]

I would download TechPowerUp GPU-Z to monitor your gpu usage.

Artifacts can happen in Firefox if you are running your gpu while browsing. Suspending the gpu tasks while browsing should make those artifacts go away. If it is firefox, then I would not worry about them. I believe It is just firefox not refreshing the video memory that the gpu utilized. I don't know about artifacts in other applications.

I would check your cpu tasks. Boinc has a tendency of running to many cpu tasks and taking away the resources need to complete gpu tasks. It is not likely with Milkyway as it requires very little cpu usage, but it could happen. I like watch GPU-Z and suspend the cpu tasks to see if it makes a difference in gpu utilization. If it does, then I know that I need to do some modifications to my configurations to scale back my cpu usage to make sure I have enough left over for the gpu.

I hope this helps.

 

[freelian]

Also, make sure the temperature is not to high on one of the cards. With two GPUs, one will often reduce the airflow of another. Tools like GPU-Z will give GPU temperatures. You may need additional cooling if one of the GPUs is running to hot.

 

[freelian]

You are probably using around 95 watts (assuming it is similar to the 760) for the CPU and 330 watts for the GPUs plus a few watts more for everything else. I would guess you are using less than 500 watts.

I would want a larger power supply as your power supply is not going to be very efficient and you are pushing it very close to the limit.

 

[freelian]

There are ways to check to see how many watts you are really using. For me, my UPS displays how many watts my computer is using. Some of CPU software monitoring tools will provide watts used. If your real usage is only around 400 watts and not something like 499 watts, you are probably ok with your current psu. The numbers I gave you are not exact. For example, I have a CPU that comes in a 95 watt package, but it only pulls around 70 watts under a full load.

 

[funkybomber]

Thanks for your time freelian,

Well it seems to have been working ok for 2-3 hours now, without me having changed anything.
As I'm writing this both GPUs are at full load executing tasks efficiently, and 2 of my 4 CPU cores doing World Community Grid tasks.

According to GPU-Z the temperatures are:

74 C for my 5850 under full load
73 C for my 7870XT under full load

So the temperatures should not be an issue.

If you have any software to suggest in order to measure my Watt consumption, I'm willing to give it a go.

 

[freelian]

I am glad to hear that it is running stable. The temperatures look good. I like to keep mine below 70, but I've heard of people running much higher temperatures without issues.

Cpuid has a tool called hwmonitor that will monitor various settings on your system like temperature and energy usage. I would be very careful when downloading and installing software from them. The last time I downloaded software from them through cnet, it had what looked like terms and conditions, but it was to actually there to install adware on your system. Just read everything and hit decline for the adware. Hwmonitor can show how much power the CPU is using and I think it shows the motherboard as well. The gpu-z you are using shows the percentage of tdp being used for each card. I think tdp is 150 watts for 5850 and around 180 watts for the 7870 so you should be able to get a rough idea from that.

The most accurate way is to use real hardware like a ups reading or a reading from a kill-a-watt device, but hopefully the software will give a good enough idea.

Good Luck.

 

[DarkRyder]

Quote:

Originally Posted by funkybomber

Thanks for your time freelian,

Well it seems to have been working ok for 2-3 hours now, without me having changed anything.
As I'm writing this both GPUs are at full load executing tasks efficiently, and 2 of my 4 CPU cores doing World Community Grid tasks.

According to GPU-Z the temperatures are:

74 C for my 5850 under full load
73 C for my 7870XT under full load

So the temperatures should not be an issue.

If you have any software to suggest in order to measure my Watt consumption, I'm willing to give it a go.

its not always the gpu temps, make sure the vrms arent getting too hot. 120c is max for vrms, i wouldnt let them get to 110c. if they are under 100c they are fine.

 

[Finrond]

Kill-a-watt power meter was one of the best things I've purchased, and they are pretty cheap. $25 or so IIRC.

 

[Tex1954]

Quote:

Originally Posted by Finrond

Kill-a-watt power meter was one of the best things I've purchased, and they are pretty cheap. $25 or so IIRC.

You got that right! I use mine all the time! Works great for tuning and such...

http://www.amazon.com/P3-International-P4460-Electricity-Monitor/dp/B000RGF29Q/ref=sr_1_1/175-2962386-1638558?ie=UTF8&qid=1377295080&sr=8-1&keywords=kill-o-watt+meter

 

[DarkRyder]

i use mine on occasion to see the different draws from my systems after hardware changes.